184 



Fishery Bulletin 102(1) 



North Pacific Swordfish 



o 



o 

 eg 



Western Pacific Bigeye 



o 



CM 



I 



o 

 I 



Western Pacific 

 Bigeye 



Western Pacific Distant 



South Pacific 

 Yellowfin 



South Pacific Bluefin 



— I — 

 140 



T 



T 



160 



220 



180 200 



Longitude (degrees) 



Figure 3 



Geographical distribution of the observer data analyzed for each fishery. 



240 



the expected number of animals per hook. For each longline 

 segment (j) within each operation (£), we link jr to a linear 

 predictor ( ?; ( ) through the equation 



rjj is then modeled as a function of soak time: 



r?y = ft+AT y , (5) 



where T tJ = the hook's soak time (decimal hours) in long- 

 line segment j; 



P = the intercept; and 



/3j = the slope coefficient, which we term the "soak 

 time coefficient." 



Modeling the probability of a catch on each individual 

 hook would result in large numbers of zero observations 

 and thus test the limits of current computer performance. 

 Therefore we aggregated hooks and catches into hourly 

 segments for each longline operation. 



We assumed that each longline segment had the same 

 configuration and that the probability of capture was the 

 same for each segment within a longline operation. The 

 assumption may be violated where segments pass through 

 different water masses or where they differ in depth profile 

 or baits. Saturation of segments with animals will also al- 

 ter the capture probability between segments. The effects 



of water masses, depth profiles, baits, and gear saturation 

 were not analyzed in the present study. 



Capture probability may also vary through the differen- 

 tial exposure of segments to the diurnal cycle of night and 

 day. The addition of dawn and dusk as fixed effects allowed 

 us to model diurnal influences on catch rates. 



Fixed effects To explore factors that might affect the rela- 

 tionship between soak time and catch rate, we added four 

 fixed effects to the logit model: year, season, and, as men- 

 tioned above, whether the segment was available at dawn 

 or dusk. A full model without interaction terms would be 



iu = A. + /Wj + AA> + PAi + PAj + P* Y u + °- 



where 7 1 , = the hook's soak time (decimal hours) in long- 

 line segment j; 



A t = an indicator of whether the hook was exposed 

 to a dawn period; 



P = an indicator of whether the hook was exposed 

 to a dusk period; 



S: , = the season (winter or summer); 



Y- = the year; 



O i = the random effect for operation that we mod- 

 eled as an independent and normally distrib- 

 uted variable (see "Random effects" section); 

 and 

 )3 -/3 4 are parameters (fixed effects) to be estimated. 

 We refer to fi x as the "soak time coefficient." 



